What Is Claude Code? A Beginner's Guide

You've probably used an AI chatbot to write some code: you ask, it answers with a block of code, you copy it somewhere and hope it works. Claude Code is different. It doesn't just suggest code — it does the work: it reads your project, edits the files, runs the commands, sees the errors, and fixes them.

That makes it far more capable, and it means a few things are worth understanding before you start.

This describes Claude Code as of September 2026. Anthropic's documentation has current details.

Chatbot vs agent

The difference is the key to understanding Claude Code.

A chatbot only talks. It can't see your files unless you paste them in, can't run your app, and can't check whether its suggestion works. You're the one doing all the copying, pasting, and testing.

An agent can take actions. Claude Code can:

Read your whole project to understand how it fits together
Edit and create files directly
Run commands — install packages, start your app, run tests
See the results — including errors — and adjust
Use git — make commits, create branches, open pull requests

So instead of "here's some code, good luck," you get: it made the change, ran it, hit an error, fixed it, and ran the tests again. More in What Is an AI Coding Agent?

Where you can use it

Claude Code runs in several places:

In a terminal — the original version. You type claude in your project folder and talk to it there. (New to terminals? The Terminal for Complete Beginners.)
In code editors — extensions for VS Code and JetBrains IDEs.
In a desktop app — Anthropic's Claude desktop app.
On the web — running in the cloud from a browser.

They're the same agent with different windows onto it.

What it costs

Claude Code is included with Anthropic's paid Claude plans (Pro, Max, Team, and Enterprise), with usage limits that depend on the plan. You can also use it with an API key, paying per use. Which is better depends on how much you use it — we compare them in Claude Pro vs Max vs API Key for Claude Code.

A first session

Once it's installed (follow Anthropic's current install instructions for your system) and you've signed in:

Open your project folder in a terminal and run claude.
Start by asking it to explain rather than change: "What does this project do? How is it organised?" It's a great way to understand code you — or another AI — wrote.
Then ask for something small and specific: "Add a 'forgot password' link under the login button that goes to /reset."
Watch what it does. It will show you each change and ask permission before running commands.

It will ask permission — and that's good

In Manual mode, Claude Code asks before it edits files or runs commands. You'll see prompts like "Run npm install? Yes / No." On Pro, Max, and Team plans, recent versions start in auto mode instead, where a safety classifier reviews risky actions and fewer prompts appear. Either way, press Shift+Tab to switch modes — and while you're learning, Manual mode is worth using: seeing each action is how you stay in control and learn what it does. (Claude Code Permission Modes explains each mode.)

As you get comfortable, you can pre-approve safe, routine commands (like running tests) so it asks less. Hold off on switching prompts off entirely until you understand the risks — an agent that can run any command can also delete things. Running Claude Code Unattended explains how to loosen things safely.

Tips for getting good results

Be specific. "Make it better" produces random changes. "Make the sign-up button bigger and green" produces the change you want. (How to Write Good Prompts for AI Coding Tools.)
One thing at a time. Small requests are easier to check and easier to undo.
Plan big changes first. For anything large, use plan mode (press Shift+Tab) so it proposes an approach before touching anything. (Claude Code Plan Mode.)
Save before big changes. Commit with git first, so you can go back. (Git and GitHub for Beginners.)
Give it a CLAUDE.md file. A short file in your project telling it how things work and what never to do. It reads this every session. (How to Write a CLAUDE.md.)
Check the result yourself. Click through the app. AI agents are confident even when wrong.

Things to be careful about

It runs on your computer by default, with access to your files. Keep secrets and important data you don't want it touching out of reach — or run it somewhere isolated.
It can reach whatever your app can reach. If your project has the password to your live database, so does the agent. Use a separate database for development. (How to Stop an AI Agent From Deleting Your Production Database.)
It doesn't remember between sessions the way a person does. Conversations can be resumed, but the most reliable memory is what's written in files. (How to Resume a Claude Code Session.)

The short version

Claude Code is an AI that doesn't just write code — it builds, runs, and checks it. Start small, let it ask permission, save your work with git before big changes, and check what it did. That's enough to get real value from day one.

EasySpawn runs Claude Code in a persistent, isolated cloud workspace with your app's database and live URL alongside it — you bring your own Claude subscription, and reach it from any device. See how it works or join the waitlist.